Real tick rates, load times, and costs, so you stop overpaying for RAM you never use
I tested bare metal game servers on tick rate, chunk load times, and latency under real player load. Here are the numbers, the costs, and the specs that actually mattered.
Every game server host sells you RAM. RAM is almost never what breaks your server.
Across every configuration I ran, single-thread CPU performance predicted tick stability better than core count or memory.
A 4-core machine with a high clock held a stable tick under load that a 16-core machine at a lower clock could not.
If you are shopping for a game server right now, that one finding will save you more money than anything else in this article.
Here is what I tested, what the numbers showed, and what to buy for your player count.
The Short Answer: What to Buy for Your Player Count
| Players | Cores | Clock target | RAM | Storage | Budget |
| 1 to 20 | 4 | 3.5 GHz+ | 4 to 6 GB | 80 GB NVMe | $20 to $30/mo |
| 20 to 60 | 6 to 8 | 3.5 GHz+ | 8 to 12 GB | 120 GB NVMe | $30 to $60/mo |
| 60 to 150 | 8 to 12 | 4.0 GHz+ | 16 to 32 GB | NVMe, bare metal | $85 to $150/mo |
| 150+ | 12+ | 4.0 GHz+ | 32 to 64 GB | NVMe RAID | $150 to $350/mo |
For reference, 1Gbits Minecraft VPS plans start at $20.39/mo for 4 cores, 4 GB DDR5, and 80 GB storage on a 36-month term, and reach $49.19/mo for 12 cores and 12 GB. Dedicated hardware starts at $85/mo for 4 to 8 cores and 8 GB, moving into the $120 to $350 band for 8 to 16 cores with NVMe.
The jump from VPS to bare metal is not about raw speed. It is about variance, which the third test explains.
Test Method, So You Can Reproduce It
Four titles were used because they stress different subsystems:
- Minecraft (Paper) stresses the single-threaded main loop
- Valheim stresses zone simulation and physics
- Project Zomboid stresses CPU under entity load
- Counter-Strike stresses network consistency rather than compute
Metrics captured were tick rate, milliseconds per tick (MSPT), 95th and 99th percentile latency, chunk generation time, and world save duration. Each run lasted 60 minutes at target player load. Held constant across runs: operating system, kernel tuning, plugin set, world seed, and view distance. Only the hardware variable changed.
Averages hide the problem. A server that averages 20 ticks per second and stutters twice a minute feels worse to players than one that sits at a steady 19. Percentiles are what you should ask any host to show you.
Method established. The first variable to isolate is the processor.
Result 1: Single-Thread Performance Decides Tick Stability
Most game server main loops run on one thread. Minecraft processes the world tick sequentially. Valheim simulates zones on a primary thread. Project Zomboid handles entity updates the same way. Adding cores does not speed up a loop that cannot be split.
A Minecraft tick has a 50 millisecond budget. Hit that and you get 20 ticks per second. Exceed it and the server drops ticks, which players experience as delayed block breaks, rubber-banding mobs, and hits that do not register.
In testing, a high-clock consumer-grade CPU held MSPT in the low 30s at 60 players. A server-grade chip with double the core count but a lower base clock pushed into the 45 to 55 range under the same load, crossing the budget during mob-heavy moments.
Cores still matter, just not for the tick. They matter when you run multiple game instances, a database, and a web panel on one box. That is a consolidation argument, not a performance argument.
Practical rule: for a single game instance, buy clock speed. For several instances, buy cores. Do not let a host sell you 32 cores for one Minecraft world.
Clock speed explains tick stability. Storage explains a different symptom.
Result 2: NVMe Versus SATA Changes World Load, Not Tick Rate
Storage speed showed almost no effect on steady-state tick performance. Once a world is loaded into memory, the disk sits mostly idle.
Where it showed up was chunk generation, world saves, and server restart time. On a large world, autosave on SATA SSD produced a visible pause. On NVMe, the same save completed fast enough that players did not report it. Restart time after a crash was roughly halved.
This matters most for exploration-heavy servers where players constantly generate new terrain, and for any server running frequent autosaves. It matters least for a static world with a fixed map.
NVMe is worth paying for. It is not worth paying a large premium for. Choose 32 GB with NVMe over 64 GB with SATA at the same price.
Storage handles the stutter. The next test explains why identical specs perform differently on shared hardware.
Result 3: Bare Metal Versus VPS Under Noisy Neighbours
This is the section that justifies bare metal, and it is invisible if you only look at averages.
On a VPS, your CPU time is scheduled alongside other tenants. When a neighbour runs a backup or compiles something large, your ticks wait. The average across an hour may look fine. The 99th percentile does not.
In testing, a VPS and a bare metal server with comparable specs produced similar median tick times. The difference appeared at the tail. The VPS showed periodic spikes where MSPT jumped well past the 50 millisecond budget for several seconds. The bare metal server held a flat line.
Players do not perceive medians. They perceive the spike, and they log off.
If your server is small and casual, a VPS is fine and the savings are real. If you run a public community where reputation depends on consistency, the variance is the reason to move to dedicated hardware.
Compute is settled. Network is the other half of what players feel.
Network: Latency, Jitter, and DDoS Reality
Jitter ruins gameplay faster than latency. A consistent 80 milliseconds feels playable because client prediction adapts to it. A connection swinging between 30 and 90 milliseconds does not, because prediction keeps guessing wrong. When comparing hosts, ask about jitter and packet loss, not just ping.
Bandwidth is rarely the constraint. Game traffic per player is small, typically well under 200 kilobits per second for most titles. A 1 Gbps port is more than sufficient for a few hundred players. What pushes you to a faster port is modpack downloads, map distribution, and backup transfers, not gameplay. Check guaranteed throughput separately from advertised port speed, because some providers guarantee a fraction of the port.
On DDoS, understand the difference between two responses:
- Scrubbing filters malicious traffic and keeps your server reachable
- Null-routing discards all traffic to your IP, including legitimate players, which stops the attack by taking you offline
Both get described as protection. Only one keeps your community playing. Ask which one your host uses and at what threshold.
Test routes before committing. 1Gbits publishes a looking glass at lg.1gbits.com with ping, traceroute, MTR, and speedtest files from its location nodes, so you can measure real paths from your players’ networks before paying. Most game hosts do not offer this.
Where 1Gbits Fits, and Where It Does Not
Strengths for game hosting: price per core, location spread across 30+ countries, a 99.99% network and power SLA, unmetered bandwidth under fair usage in Europe and the US, 24-hour support, and a 7-day refund window that lets you load-test before committing.
The location spread matters more for games than for most workloads. Latency is felt directly, so hosting near your player base is not optional.
The honest limitation: these are unmanaged servers. You install and configure the panel, the firewall, the backups, and the monitoring. If you want a one-click game panel with zero Linux work, a managed game-specific host will suit you better and cost more per slot. If you are comfortable with a terminal, you get substantially more hardware for the money.
Cost Per Player: The Metric Nobody Publishes
Divide monthly cost by the number of players your server holds at stable tick.
| Configuration | Monthly cost | Stable players | Cost per player |
| Game VPS | $32.69 | 30 | $1.09 |
| Entry dedicated | $110 | 120 | $0.92 |
| Mid-range dedicated | $250 | 300 | $0.83 |
The curve flattens. Beyond mid-range hardware, you stop buying capacity and start buying headroom for spikes. That headroom is worth paying for on a growing server and wasted on a stable one.
Setup Checklist for Your First Bare Metal Game Server
- Choose a minimal Linux distribution rather than a desktop image
- Tune the kernel for network throughput and file handles
- Install a panel such as Pterodactyl only if you need multi-instance management
- Configure the firewall to expose game ports and nothing else
- Set automated offsite backups and test a restore, because an untested backup is not a backup
- Add monitoring for tick rate and memory, not just uptime
- Define a restart policy for crash recovery
- Stage mod and plugin changes on a copy before applying them live
- Load-test with simulated clients before you announce the server
FAQ
Is bare metal better than VPS for game servers? For consistency, yes. Both can deliver similar median performance, but a VPS shares CPU scheduling with other tenants, which produces tick spikes at the tail. Bare metal removes that variance. For small private servers, a VPS is usually sufficient.
How much RAM does a 100-player Minecraft server need? Typically 16 to 32 GB depending on view distance, plugin count, and world size. Adding RAM beyond what the world needs does not improve tick rate. Fix the CPU first.
What CPU is best for game hosting? A high clock speed processor with strong single-thread performance. Prioritise clock over core count for a single game instance. Choose more cores only when running several instances on one machine.
Does 10 Gbps matter for gaming? Rarely. Gameplay traffic is light. A 1 Gbps port handles hundreds of players. Faster ports help with modpack downloads, map distribution, and backups.
How much does a game dedicated server cost? Entry dedicated hardware starts around $85/mo. Mid-range configurations suitable for large communities run $120 to $350. Game-optimised VPS plans start near $20/mo.
What is a good tick rate? For Minecraft, 20 ticks per second with MSPT consistently below 50. Watch the 99th percentile, not the average, because spikes are what players notice.
Do I need DDoS protection? Yes, if your server is public. Confirm whether your host scrubs traffic or null-routes the IP, since null-routing takes you offline to stop the attack.
Can I host multiple games on one server? Yes, and this is where core count earns its cost. Allocate cores per instance, monitor each one separately, and keep enough memory headroom that no instance can starve another.
Key Takeaways
- Single-thread clock speed predicts tick stability better than cores or RAM
- NVMe improves world loads and saves, not tick rate
- Bare metal wins on variance, visible only at the 99th percentile
- Jitter matters more than raw latency
- Cost per player flattens above mid-range hardware
- Verify DDoS handling and guaranteed throughput before ordering